Integration of ForceTorqueControl (FTC) for fine sanding

    Hi guys, i've been a lurker on this forum so far, but have now found the need to reach out to the community with a question.

    I am looking to add KUKA ForceTorqueControl to our KR10 R1100(KSS 8.5.8) for the purpose of sanding and polishing.

    I'm considering a 6-axis ATI (or alternatives) Force Torque Sensor over ETHERCAT to our KR C4.

    This question is based on a demo we did where:

    -6 PTP locations were defined on key spots on the surface

    -The surface was complex, double curved with multiple bumps and humps

    -Force/Torque control with compliance was used to keep the sanding pad perpendicular to the surface at all times while moving to these (PTP) locations

    -The target force was only about 2N and was run in TTS mode(Tool base Technology System, $TOOL)

    -> Results were awesome! Very even and precise sanding job without ruining some of the fine details in the surface. All the area's were touched equally and details were left in place. A softer type sanding interface pad was used.


    Now the part where I'm wondering about is the capabilities and the integration of such a system with KUKA and a ForceTorqueControl Package.

    I short I want to know what I am getting into and if this is a project that the package TorqueForceControl is suited for. The cobot package was plug and play, what can the kuka package do out of the box? Is a lot of tweaking in RSI needed to get it to work like described?

    I've seen a lot of assembly usecases for the FTC package, but not so really anything with fine, dynamic movement and finesse. My usual sources are blank when it comes to this subject, which is why I am reaching out on this forum.

    My experience is on an intermediate level but we have contacts with skilled integrators that could lend a hand. None of them have any prior experience with Force/Torque Control(-ers) though.


    Thanks for any insight you can share :red_heart:

  • Go to Best Answer

    if you have something that you you spend time on developing, something that works well and you can control, why not stick with it? you get get manuals on FTC from Kuka Xpert portal. then you can see how it works, what is required and what the capabilities are.

    1) read pinned topic: READ FIRST...

    2) if you have an issue with robot, post question in the correct forum section... do NOT contact me directly

    3) read 1 and 2

    How it looks from the manual it should be possible to do. I can set a target load in a reference coordinate system, work in TTS mode and use the compliance parameters to achieve this in theory. This is however may be far from a working system.

    The fact that I don’t see any applications like what I want on a KUKA with FTC makes me wonder if the system’s performance would be in any way sufficient.

    As you said, if I can control it, spend time on development and it works well I know i will stick with it. I’m just not sure it will work well yet. I was hoping to find someone with a similar application here :smiling_face:

    Quote from Retsok5

    This question is based on a demo we did where:

    -6 PTP locations were defined on key spots on the surface

    -The surface was complex, double curved with multiple bumps and humps

    -Force/Torque control with compliance was used to keep the sanding pad perpendicular to the surface at all times while moving to these (PTP) locations

    -The target force was only about 2N and was run in TTS mode(Tool base Technology System, $TOOL)

    Quote from Retsok5

    Now the part where I'm wondering about is the capabilities and the integration of such a system with KUKA and a ForceTorqueControl Package.

    ...I'm confused. If you already did it once, then what's the question?

    I see a very important bit of information has been left out by accident, my apologies..

    We did the demo on a cobot, not on a KUKA. The cobot had a simplified interface and operating system to these types of compliance and force control tasks.

    I am wondering if a KUKA based system can be set up to do the same

    KUKA has cobot and it is called LBR iiwa. and it is by far the most sensitive cobot on the market based on what i have seen so far. and it is available for 14kg payload which is more than your current target is rated for.

    Quote from Retsok5

    The fact that I don’t see any applications like what I want ...

    .... I was hoping to find someone with a similar application here.

    maybe when you find some time, try sharing your ideas. so far all i see in this thread is very secretive.

    no mention that cobot was used, what the cobot type was, what type of sensing was used, what exactly is expected from FTC that documentation does not cover... i mean throw us a bone...

    and if it involves something that require discussion behind blast doors of some underground chamber in a military compound like Area51, then why even mention it in a public place like this robot forum?

    1) read pinned topic: READ FIRST...

    2) if you have an issue with robot, post question in the correct forum section... do NOT contact me directly

    3) read 1 and 2

    • Best Answer
    Quote from Retsok5

    We did the demo on a cobot, not on a KUKA. The cobot had a simplified interface and operating system to these types of compliance and force control tasks.

    I am wondering if a KUKA based system can be set up to do the same

    Probably yes. There are always caveats, of course. But with a 6DOF load cell mounted between the A6 flange and the sanding tool, it should be possible to get similar performance. Of course, adding FTC to a KUKA robot does not make it a Cobot, so the safety considerations will be those of a "big iron" robot.

    Items to keep in mind when doing FTC:

    1. Distance between the load cell and the point-of-contact needs to be minimal -- the longer the "lever arm" is, the more any sensing errors will be amplified
    2. The mounting of the tool to the load cell needs to be rigid. Any compliance between the point-of-contact and the load cell can cause erroneous readings. The exception to this rule is when compliance is deliberately added along/around a specific axis, usually with robust springs or damping, to avoid "step" transients doing a move-to-contact operation. In your case, you might want to add a short-stroke, stiffly-sprung compliance parallel to the rotation axis of the sanding head, but keep the tool-to-load-cell relationship rigid in all other respects.
    3. Some degree of filtering and smoothing will be required. In KUKA FTC, there are variables available for this. Tuning will need to be done experimentally. FTC will gladly overdrive the robot, so it's up to you, while tuning your FTC application, to control the tuning, speed, and limits to avoid over-driving.
    4. Load cells aren't perfect, and can "drift" over time. This drift is usually an offset in the zero point, not in the force/signal ratio. So typical practice is to get the robot into the "attack" pose, just shy of making contact, then taring the load cell before moving to contact. This helps isolate the contact forces from the gravity forces.
    Quote from SkyeFire

    Probably yes. There are always caveats, of course. But with a 6DOF load cell mounted between the A6 flange and the sanding tool, it should be possible to get similar performance. Of course, adding FTC to a KUKA robot does not make it a Cobot, so the safety considerations will be those of a "big iron" robot.

    ....

    Skyefire, this is already very insightful. Many thanks!

    I am aware of the safety factor when comparing the two robots. This does not pose an issue.

    Your points are all very clear. It’s good to so know what some of the physical considerations are of such a system are. On the software side, in your third point I sense quite a challenge to get reliable performance..

    Am I correct to think that the configuration of filtering and smoothing is done in RSI?

    Quote from panic mode

    maybe when you find some time, try sharing your ideas. so far all i see in this thread is very secretive.

    no mention that cobot was used, what the cobot type was, what type of sensing was used, what exactly is expected from FTC that documentation does not cover... i mean throw us a bone...

    I see your point. This is not my intention.

    Some elaboration and hopefully clarification:

    To get a bit more concrete I’ve taken some time to elaborate on the usecase:


    Consider a pneumatic random orbital sander with a 75mm / 3" soft backing pad. We want to sand/polish our guitars with it. The instruments are high end and have a lot of details in the curvature of the body. This link https://imgur.com/KSHKX1m shows the curvature of the model and an example program with how I’d like to see it move over the surface. Notice the different orientation of the sanding pad in each scenario.


    I define the toolpath using our CAM software. This output toolpath will contain XYZ and ABC values for every segment. I tested this on our KUKA without FTC and found that the result wasn't great: the applied load was uneven which resulted in some sections being (over)done and others being untouched. A perfect case for the application of FTC you'd think..


    As I talked about, we then did the test on the Doosan Cobot and the result was great. I was astonished as to how little programming was required to get these results. Part of it comes from the constant force that is exerted and the other part comes from the compliance controller. It uses this to stay perpendicular to the surface at all times. The constant force could be achieved via passive concepts as well, and so I am mainly looking to get the compliance functionality.

    From the FTC manual I gather that the package can be used to achieve the same functionality with a KUKA. Both systems make use of load cells.

    How I think it would look in terms of force and compliance:

    • Setpoint values: Fx=none, Fy=none, Fz= 2N, Tx= 0, Ty=0 Ty=none
    • Correction DOF’s are (in $TOOL): angles A, B and C plus along the Z axis (values TBD)

    From what I can see this means that the system should be able use compliance to stay perpendicular to the surface (target: Fz=2N, Tx=0 and Ty=0) by making use of active angle and z-axis compensation.

    In theory a FTC based KUKA system should be about equally well equipped as the Doosan cobot. However, like I said. I haven’t seen a similar application anywhere. I imagine this requires a bit more (software)engineering than the standard applications as described in the ForceTorqueControl manual as they seem less complex. This is where my question originates from.

    Skyefire already mentioned it should be possible to get a similar performance with a 6DOF on a KUKA, but maybe this elaboration changes things..

    Is this a realistic usecase for this package(software and hardware)?

    Quote from Retsok5

    Am I correct to think that the configuration of filtering and smoothing is done in RSI?

    Not necessarily. FTC is a sort of "easy mode" layer atop RSI, and includes a tuning variable for the reaction speed/strength of each axis, in mm/sec/N (or deg/sec/Nm for torque). For a lot of applications, this is sufficient.

    For more complex situations, it's possible to perform certain customizations to an FTC program using the RSI Visual editor, since installing FTC brings along RSI as an included dependency. This can get into deep water pretty quickly, but making minor changes (like adding a filter to the load cell signals to reject the high-frequency vibrations of your sander) are doable.

    Rule of thumb: try FTC first, and save RSI for when you really need it.

    Quote from Retsok5

    How I think it would look in terms of force and compliance:

    Setpoint values: Fx=none, Fy=none, Fz= 2N, Tx= 0, Ty=0 Ty=none
    Correction DOF’s are (in $TOOL): angles A, B and C plus along the Z axis (values TBD)

    That seems like something FTC should be able to handle well enough. Your TCP will probably need to be at the center of the sanding disk.

    Quote from Retsok5

    In theory a FTC based KUKA system should be about equally well equipped as the Doosan cobot. However, like I said. I haven’t seen a similar application anywhere. I imagine this requires a bit more (software)engineering than the standard applications as described in the ForceTorqueControl manual as they seem less complex. This is where my question originates from.

    I don't see any fundamental reason why a KUKAbot with FTC would be any less capable of doing this application than a Cobot, although I'm not familiar with the Doosan so it may have capabilities I'm not aware of.

    That brings up a question: if you built a demo with the Doosan that worked, why do you want to switch to a KUKAbot? What do you want the KUKA to do differently compared to the Doosan demo?

    Hi Retsok5,

    Did you manage to implement this and if the answer is yes, would you share your experiences or maybe some problems that you managed to solve that are not in the Manual of Force Torque Control?

    We are thinking of getting this package so any advice would be appreciated

    Thank you in advanced,

    Regards to all

    Some issues came up that required my undivided atention so this project has been put to the side for the last few weeks. My appologies for the silence on my part. I will pick this up soon again.

    However, we have yet to make the decision if we want to move in this direction or not. If we decide to move in that direction I'd still need to acquire the FTC packages and do some test before I can get back with any sort of results.

    Thanks for your patience and my apologies for the delay

Advertising from our partners